Fix sendmmsg buffer overrun causing EFAULT when count > 1024

[Copilot]

The -g flag (sendmmsg) fails with "Bad address" (EFAULT) when the engine attempts to send more than 1024 packets at once.

Root Cause

send_packets_using_sendmmsg() declares struct mmsghdr mmsgs[1024] on the stack. The initialization loop caps at 1024 elements, but the sendmmsg syscall was invoked with the full count parameter:

for (i = 0; i < count && i < sizeof(mmsgs) / sizeof(mmsgs[0]); ++i)
    // Initialize mmsgs[i]

s = sendmmsg(fd, mmsgs, count, 0);  // count may exceed 1024

When count > 1024, the kernel accesses uninitialized memory beyond the array bounds.

Changes

Pass the actual number of initialized messages to sendmmsg:

s = sendmmsg(fd, mmsgs, i, 0);
if (s < (int) i)

Where i = min(count, 1024) after the loop.

Original prompt

---

This section details on the original issue you should resolve

<issue_title>sendmmsg() and recvmmsg() fail with bad address</issue_title>
<issue_description>### Description

When trying to transmit a 8GB file using the -j and -g parameters, the connection fails. The handshake seems to complete according to the debug logs. The following errormessage is printed:

12:56:42.361 [WARN] sendmmsg failed: Bad address
12:56:42.361 [WARN] engine: error sending packet for regular connection CF436FAFAB812484 - close it; src: 0.0.0.0:47305; dst: 10.0.0.2:4433; errno: 14

When using the exact same configurations but omitting the -j -g in both commmands, everything works as it should.
Using just one of the two options or on just one endpoint leads to the same error.

Environment:

• Two machines with Debian Bullseye (Linux 5.10.0-8-amd64, x86_64)
• One machine running the http_client, the other one running the http_server
• Machines connected with direct 10G link. IP addresses 10.0.0.{1,2}, hostname server in /etc/hosts of the client
• Client started with ./http_client -p file -7 download -H server -s server:4433/file -G logs -j -g
• Server started with ./http_server -s 10.0.0.2:4433 -r www -c server,certs/cert.pem,certs/priv.key -G logs -j -g

Are there any known issues with those options?
Thank you for your help!</issue_description>

Comments on the Issue (you are @copilot in this section)

@dtikhonov Thank you for the bug report. We will get to this at some point.

If you could recheck this (it's been years), it would be great.</comment_new>
<comment_new>@dtikhonov
Thank you for the detailed analysis, @pkropachev!

Do you want to take a stab at fixing this?</comment_new>

• Fixes sendmmsg() and recvmmsg() fail with bad address #434

---

✨ Let Copilot coding agent set things up for you — coding agent works faster and does higher quality work when set up for your repo.